# Knowledge-based Agent Reasoning on KARLBench 1.0 (test) Source: [Knowledge-based Agent Reasoning on KARLBench 1.0 (test)]() Collection index: [llms.txt]() Data updated: 2026-03-06T09:31:11.363Z Evaluation of model performance on Knowledge-based Agent Reasoning tasks using the test split of the KARLBench 1.0 dataset, reporting scores across specialized components like BrowseComp-Plus, FinanceBench, and overall performance. A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage. | Scope | Recorded value | | --- | --- | | Task | Knowledge-based Agent Reasoning | | Dataset (including any stated split) | KARLBench 1.0 (test) | | Metric | BrowseComp-Plus Score | | Unit | PERCENT | | Direction | Higher is better | | Results loaded | 17 of 17 | | Coverage | All results currently returned for this benchmark | Evaluation settings may differ between results. The original papers and recorded configurations are needed to establish comparability. A shared metric name alone does not establish an identical protocol. ## Metrics - [BrowseComp-Plus Score](): Higher is better; unit: PERCENT. - [TREC-Biogen Score](): Higher is better; unit: PERCENT. - [FreshStack Score](): Higher is better; unit: PERCENT. - [FinanceBench Score](): Higher is better; unit: PERCENT. - [QAMPARI Score](): Higher is better; unit: PERCENT. - [PMBench Score](): Higher is better; unit: PERCENT. - [In-Dist. Accuracy](): Higher is better; unit: PERCENT. - [OOD Accuracy](): Higher is better; unit: PERCENT. - [Total Accuracy](): Higher is better; unit: PERCENT. ## Recorded results | Method | BrowseComp-Plus Score | Result date | Report paper | | --- | --- | --- | --- | | Claude 4.6 Opus | 75.9 | Not recorded | 69aa5f932d603e6de3e4197d | | KARL-BCP (VGS N = 17) | 70.4 | Not recorded | 69aa5f932d603e6de3e4197d | | KARL (par. N = 20) | 69.5 | Not recorded | 69aa5f932d603e6de3e4197d | | GPT 5 | 68.3 | Not recorded | 69aa5f932d603e6de3e4197d | | KARL (par. N = 10) | 67.5 | Not recorded | 69aa5f932d603e6de3e4197d | | Claude 4.5 Opus | 62.5 | Not recorded | 69aa5f932d603e6de3e4197d | | KARL (par. N = 3) | 62.2 | Not recorded | 69aa5f932d603e6de3e4197d | | KARL-BCP | 59.6 | Not recorded | 69aa5f932d603e6de3e4197d | | KARL | 58.5 | Not recorded | 69aa5f932d603e6de3e4197d | | Claude 4.6 Sonnet | 57.9 | Not recorded | 69aa5f932d603e6de3e4197d | | Minimax m2.5 | 56.5 | Not recorded | 69aa5f932d603e6de3e4197d | | Qwen 3.5 397B A17B | 55.8 | Not recorded | 69aa5f932d603e6de3e4197d | | Claude 4.5 Sonnet | 54.6 | Not recorded | 69aa5f932d603e6de3e4197d | | GPT 5.2 | 47.8 | Not recorded | 69aa5f932d603e6de3e4197d | | Claude 4.5 Haiku | 45.8 | Not recorded | 69aa5f932d603e6de3e4197d | | GLM 4.5 Air | 44.7 | Not recorded | 69aa5f932d603e6de3e4197d | | KARL-TREC | 42.2 | Not recorded | 69aa5f932d603e6de3e4197d | ### Claude 4.6 Opus BrowseComp-Plus Score: 75.9. Result ID: 69aa9ed56dbb7075b954ae7b. [Reporting paper]() Recorded evaluation configuration: {"reasoning effort":"best of low, medium, and high","context management":"best with/without compression"}. This record does not assert independent reproduction. ### KARL-BCP (VGS N = 17) BrowseComp-Plus Score: 70.4. Result ID: 69aa9ed56dbb7075b954ae7e. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Training Setup":"Single-Task RL","Value-Guided Search (VGS)":true,"N":"17"}. This record does not assert independent reproduction. ### KARL (par. N = 20) BrowseComp-Plus Score: 69.5. Result ID: 69aa9ed56dbb7075b954ae82. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Training Setup":"Multi-Task RL","Parallel thinking":true,"N":"20"}. This record does not assert independent reproduction. ### GPT 5 BrowseComp-Plus Score: 68.3. Result ID: 69aa9ed56dbb7075b954ae75. [Reporting paper]() Recorded evaluation configuration: {"reasoning effort":"best of low, medium, and high","context management":"best with/without compression"}. This record does not assert independent reproduction. ### KARL (par. N = 10) BrowseComp-Plus Score: 67.5. Result ID: 69aa9ed56dbb7075b954ae81. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Training Setup":"Multi-Task RL","Parallel thinking":true,"N":"10"}. This record does not assert independent reproduction. ### Claude 4.5 Opus BrowseComp-Plus Score: 62.5. Result ID: 69aa9ed56dbb7075b954ae79. [Reporting paper]() Recorded evaluation configuration: {"reasoning effort":"best of low, medium, and high","context management":"best with/without compression"}. This record does not assert independent reproduction. ### KARL (par. N = 3) BrowseComp-Plus Score: 62.2. Result ID: 69aa9ed56dbb7075b954ae80. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Training Setup":"Multi-Task RL","Parallel thinking":true,"N":"3"}. This record does not assert independent reproduction. ### KARL-BCP BrowseComp-Plus Score: 59.6. Result ID: 69aa9ed56dbb7075b954ae7d. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Training Setup":"Single-Task RL"}. This record does not assert independent reproduction. ### KARL BrowseComp-Plus Score: 58.5. Result ID: 69aa9ed56dbb7075b954ae7f. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Training Setup":"Multi-Task RL"}. This record does not assert independent reproduction. ### Claude 4.6 Sonnet BrowseComp-Plus Score: 57.9. Result ID: 69aa9ed56dbb7075b954ae7a. [Reporting paper]() Recorded evaluation configuration: {"reasoning effort":"best of low, medium, and high","context management":"best with/without compression"}. This record does not assert independent reproduction. ### Minimax m2.5 BrowseComp-Plus Score: 56.5. Result ID: 69aa9ed56dbb7075b954ae74. [Reporting paper]() Recorded evaluation configuration: {"context management":"best with/without compression"}. This record does not assert independent reproduction. ### Qwen 3.5 397B A17B BrowseComp-Plus Score: 55.8. Result ID: 69aa9ed56dbb7075b954ae73. [Reporting paper]() Recorded evaluation configuration: {"context management":"best with/without compression"}. This record does not assert independent reproduction. ### Claude 4.5 Sonnet BrowseComp-Plus Score: 54.6. Result ID: 69aa9ed56dbb7075b954ae78. [Reporting paper]() Recorded evaluation configuration: {"reasoning effort":"best of low, medium, and high","context management":"best with/without compression"}. This record does not assert independent reproduction. ### GPT 5.2 BrowseComp-Plus Score: 47.8. Result ID: 69aa9ed56dbb7075b954ae76. [Reporting paper]() Recorded evaluation configuration: {"reasoning effort":"best of low, medium, and high","context management":"best with/without compression"}. This record does not assert independent reproduction. ### Claude 4.5 Haiku BrowseComp-Plus Score: 45.8. Result ID: 69aa9ed56dbb7075b954ae77. [Reporting paper]() Recorded evaluation configuration: {"reasoning effort":"best of low, medium, and high","context management":"best with/without compression"}. This record does not assert independent reproduction. ### GLM 4.5 Air BrowseComp-Plus Score: 44.7. Result ID: 69aa9ed56dbb7075b954ae72. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"context management":"best with/without compression"}. This record does not assert independent reproduction. ### KARL-TREC BrowseComp-Plus Score: 42.2. Result ID: 69aa9ed56dbb7075b954ae7c. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Training Setup":"Single-Task RL"}. This record does not assert independent reproduction. Page 1 of 1. This is an index page, not the full collection.